From: Li Nan <linan122@huawei.com>

Commit a05b7ea03d72 ("md: avoid crash when stopping md array races
with closing other open fds.") added sync_block before stopping raid and
setting readonly. Later in commit 260fa034ef7a ("md: avoid deadlock when
dirty buffers during md_stop.") it is moved to ioctl. array_state_store()
was ignored. Add sync blockdev to array_state_store() now.

Signed-off-by: Li Nan <linan122@huawei.com>
---
 drivers/md/md.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/drivers/md/md.c b/drivers/md/md.c
index 2c793992a604..aea39598457c 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -4477,6 +4477,7 @@ array_state_store(struct mddev *mddev, const char *buf, size_t len)
 {
 	int err = 0;
 	enum array_state st = match_word(buf, array_states);
+	bool clear_md_closing = false;
 
 	/* No lock dependent actions */
 	switch (st) {
@@ -4511,6 +4512,16 @@ array_state_store(struct mddev *mddev, const char *buf, size_t len)
 		spin_unlock(&mddev->lock);
 		return err ?: len;
 	}
+
+	/* we will call set readonly or stop raid, sync blockdev */
+	if (st == clear || (mddev->pers && (st == readonly ||
+	    st == inactive || (st == read_auto && md_is_rdwr(mddev))))) {
+		err = mddev_sync_blockdev(mddev);
+		if (err)
+			return err;
+		clear_md_closing = true;
+	}
+
 	err = mddev_lock(mddev);
 	if (err)
 		return err;
@@ -4523,6 +4534,8 @@ array_state_store(struct mddev *mddev, const char *buf, size_t len)
 		break;
 	case clear:
 		err = do_md_stop(mddev, 0, NULL);
+		if (!err)
+			clear_md_closing = false;
 		break;
 	case readonly:
 		if (mddev->pers)
@@ -4585,6 +4598,8 @@ array_state_store(struct mddev *mddev, const char *buf, size_t len)
 		sysfs_notify_dirent_safe(mddev->sysfs_state);
 	}
 	mddev_unlock(mddev);
+	if (clear_md_closing)
+		clear_bit(MD_CLOSING, &mddev->flags);
 	return err ?: len;
 }
 static struct md_sysfs_entry md_array_state =
-- 
2.39.2

On Wed, 17 Jan 2024 17:37:07 +0800
linan666@huaweicloud.com wrote:

> From: Li Nan <linan122@huawei.com>
> 
> Commit a05b7ea03d72 ("md: avoid crash when stopping md array races
> with closing other open fds.") added sync_block before stopping raid and
> setting readonly. Later in commit 260fa034ef7a ("md: avoid deadlock when
> dirty buffers during md_stop.") it is moved to ioctl. array_state_store()
> was ignored. Add sync blockdev to array_state_store() now.
> 
> Signed-off-by: Li Nan <linan122@huawei.com>
> ---
>  drivers/md/md.c | 15 +++++++++++++++
>  1 file changed, 15 insertions(+)
> 
> diff --git a/drivers/md/md.c b/drivers/md/md.c
> index 2c793992a604..aea39598457c 100644
> --- a/drivers/md/md.c
> +++ b/drivers/md/md.c
> @@ -4477,6 +4477,7 @@ array_state_store(struct mddev *mddev, const char *buf,
> size_t len) {
>  	int err = 0;
>  	enum array_state st = match_word(buf, array_states);
> +	bool clear_md_closing = false;
>  
>  	/* No lock dependent actions */
>  	switch (st) {
> @@ -4511,6 +4512,16 @@ array_state_store(struct mddev *mddev, const char
> *buf, size_t len) spin_unlock(&mddev->lock);
>  		return err ?: len;
>  	}
> +
> +	/* we will call set readonly or stop raid, sync blockdev */
> +	if (st == clear || (mddev->pers && (st == readonly ||
> +	    st == inactive || (st == read_auto && md_is_rdwr(mddev))))) {
> +		err = mddev_sync_blockdev(mddev);
> +		if (err)
> +			return err;
> +		clear_md_closing = true;
> +	}
> +

Please reorganize it a little for readability:
I think if no mddev->pers we don't need to consider sync_blockdev at all. If
personality is there we can probably check for read-write. If it is not
read-write then nothing to sync. What about that:

if (mddev->pers && md_is_rdwr(mddev) &&
    (st == clear || st == readonly || st == inactive || st == read_auto)) 

Please note that I didn't test it so please let me know if you see issue in
proposed logic.
I think that we may be able to include it in "/* No lock dependent actions */"
switch. Please consider it too:

case clear:
case readonly:
case inactive:
case read_auto:
    if(!mddev->pers || !md_is_rdwr(mddev))
          break;
    err = mddev_sync_blockdev(mddev);
    if (err)
        return err;
    clear_md_closing = true;
    
>  	err = mddev_lock(mddev);
>  	if (err)
>  		return err;
> @@ -4523,6 +4534,8 @@ array_state_store(struct mddev *mddev, const char *buf,
> size_t len) break;
>  	case clear:
>  		err = do_md_stop(mddev, 0, NULL);
> +		if (!err)
> +			clear_md_closing = false;
>  		break;
>  	case readonly:
>  		if (mddev->pers)
> @@ -4585,6 +4598,8 @@ array_state_store(struct mddev *mddev, const char *buf,
> size_t len) sysfs_notify_dirent_safe(mddev->sysfs_state);
>  	}
>  	mddev_unlock(mddev);
> +	if (clear_md_closing)
> +		clear_bit(MD_CLOSING, &mddev->flags);

Please add spaces before and after if.

>  	return err ?: len;
>  }
>  static struct md_sysfs_entry md_array_state =

Thanks,
Mariusz

在 2024/1/18 16:02, Mariusz Tkaczyk 写道:
> On Wed, 17 Jan 2024 17:37:07 +0800
> linan666@huaweicloud.com wrote:
> 
>> From: Li Nan <linan122@huawei.com>
>>
>> Commit a05b7ea03d72 ("md: avoid crash when stopping md array races
>> with closing other open fds.") added sync_block before stopping raid and
>> setting readonly. Later in commit 260fa034ef7a ("md: avoid deadlock when
>> dirty buffers during md_stop.") it is moved to ioctl. array_state_store()
>> was ignored. Add sync blockdev to array_state_store() now.
>>
>> Signed-off-by: Li Nan <linan122@huawei.com>

[...]

>> +
>> +	/* we will call set readonly or stop raid, sync blockdev */
>> +	if (st == clear || (mddev->pers && (st == readonly ||
>> +	    st == inactive || (st == read_auto && md_is_rdwr(mddev))))) {
>> +		err = mddev_sync_blockdev(mddev);
>> +		if (err)
>> +			return err;
>> +		clear_md_closing = true;
>> +	}
>> +
> 
> Please reorganize it a little for readability:
> I think if no mddev->pers we don't need to consider sync_blockdev at all. If
> personality is there we can probably check for read-write. If it is not
> read-write then nothing to sync. What about that:
> 
> if (mddev->pers && md_is_rdwr(mddev) &&
>      (st == clear || st == readonly || st == inactive || st == read_auto))
> 
> Please note that I didn't test it so please let me know if you see issue in
> proposed logic.
> I think that we may be able to include it in "/* No lock dependent actions */"
> switch. Please consider it too:
> 

Thanks for your review.
It is a really good idea. I will test and improve it.

> case clear:
> case readonly:
> case inactive:
> case read_auto:
>      if(!mddev->pers || !md_is_rdwr(mddev))
>            break;
>      err = mddev_sync_blockdev(mddev);
>      if (err)
>          return err;
>      clear_md_closing = true;
>      

-- 
Thanks,
Nan